Automatic classification of ANA HEp-2 Immunofluorescence images based on the texture features using artificial neural network

Indirect Immunfluorsece method (IFA) is one of the important laboratory procedures for the diagnosis of the autoimmune disease, but it suffers from low throughput and subjectivity due to manual interpretation. The Human Epithelial type-2 (HEp-2) pattern, such as homogeneous, speckled, centromere, Nucleolar pattern images, gives the diagnosis of different autoimmune diseases. For the current study, different patterns are obtained from the publicly available datasets A.I.D.A ((Auto- Immunity Diagnosis by Computer) project of 1000 images. The images pre-processed and features such as statistical and textural features extracted and explored to find the appropriate one for the detection and the classification of ANA HEp2 cells pattern. The paper uses the Analysis of Variance (ANOVA) for the identification of appropriate features and Artifical Neural network (ANN) for classification. The result obtained indicates that textural features are the better features in comparison with other extracted features, with the results obtained average accuracy around 92% using ANN as the classifier. The outcome thus produced is useful for the further design of cost-effective image analysis in the autoimmune diagnosis.